Add benchmarks for Subscriber and Publisher

[AlexKaravaev]

Hi everyone!

It's a PR for #33, however, I don't really know which of the parts of the library should be benchmarked at first, currently I only have written tests for simple publishing and subscribing for simple string and creating of publisher and subscriber instances.

Ideas that I have about what can be benchmarked:

• Publishing and subscribing with more heavy payload(E.g progressing increase of message size)
• Message creation

Please feel free to add some ideas

[jhdcs]

Would benchmarking the creation and sending/receiving of messages with allocated portions (such as Strings) vs trivially-copyable portions (such as i32) be of use?

Another thing that might be good to benchmark are all the various ROS 2 object creation functions: creating nodes, creating contexts, creating publishers, creating subscribers, etc. More for catching performance regressions than anything else. While these functions mostly involve external ROS 2 code, it would give us insight into the health of the external system.

[nnmm]

@AlexKaravaev Want to revive this to gather some numbers for sending messages in the same process vs between processes?

We could construct a message, and then time the duration for

• sending the message
• receiving the message on the other end, and
• sending back a very small response (e.g. std_msgs/msg/Empty)

for different message types. See also https://gitlab.com/ApexAI/performance_test